THE EVIDENCE BEHIND THE STORY

Two recordings.
Two different results.

The final stages of How it works replay two public Stanford research trials. They are separate from Neuralink’s hardware and the fictional opening conversation.

Compare intention and prediction.

Prompted sentences and recorded final predictions
TrialPromptFinal predictionGo-period duration
Trial 363I am not okayi am it3.76 s
Trial 365Where is it?where is it2.34 s

Participant T12 · June 14, 2022. These selected examples show a match and a mismatch, not an overall accuracy estimate.

What is real, and what is illustrated?

The prompted sentences, partial predictions and final predictions come from the published dataset. The character, dialogue, room, implant placement and signal animation are educational illustrations.

Audio is browser speech synthesis, not T12’s recorded or reconstructed voice. The replay does not run a decoder live or measure Neuralink’s performance.

Research details and recording sites

Source: Willett et al. (2023), A high-performance speech neuroprosthesis; Dryad sentences dataset, selected file t12.2022.06.14_sentences.mat. The selected go periods contain 188 and 117 time bins of 20 ms, respectively.

The study recorded arrays in areas 6v and 44; speech-decoding analyses used 6v recordings. These research locations do not specify Neuralink thread placement. The interactive scene is not an anatomical placement map.

Read the paper ↗ · Open the dataset ↗

Explore How it works →